Long Short-Term Memory Graph Convolutional Network for Network-Wide Traffic Prediction

As a key part of intelligent transportation systems, traffic flow prediction faces the challenge of inaccurate long- and short-term prediction. Traditional statistical modeling, temporal analysis, and graph convolution network (GCN) alone cannot handle both temporal and spatial dependencies. In this paper, a traffic flow prediction method based on long- and short-term memory graph convolutional networks (LSTM-GCN) is proposed. The real-world traffic flow data sets are used for the experiments, and the results show that the RMSE error of our proposed LSTM-GCN model is reduced by about 41.8%, and the accuracy is improved by 4.2%, compared with the SVR models. The accuracy of the LSTM-GCN model is higher than that of a single-factor model, indicating that the LSTM-GCN model can capture spatial and temporal features from traffic data.